Nvidia Widens Korea AI Push With Science Ministry Pact and Samsung, LG, Kakao Projects

The package goes beyond selling GPUs. It aims to tie Korean chip designers, model builders and manufacturers more closely to Nvidia's software and interconnect technology, at a time when Seoul is spending heavily to build AI capacity it controls. The posts also gave concrete numbers for Samsung Electronics' engineering work and named new partners in autonomous driving, robotics and agent security.

Under the ministry agreement, Korean companies designing NPUs and other custom processors will be able to explore joining the NVLink Fusion ecosystem, Nvidia's programme for connecting third-party chips to its rack-scale systems. Nvidia will also run training and certification programmes, hackathons and technical seminars intended to help Korean users get more out of their GPUs. A separate deal with SK Telecom and chip designer Rebellions, also backed by the ministry, will use NVLink Fusion to build sovereign AI infrastructure for Korean models.

Through Korea's AI for All initiative, Nvidia said it will work with three selected consortia and four leading Korean foundation model builders, using its Dynamo inference software to scale their models. The company did not name the consortia. It is releasing Dynamo deployment recipes, tuned for its B200 GPUs, for four Korean open models: SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2 and Motif Technologies' Motif-3. Nvidia said the recipes deliver up to twice the performance without further tuning. Dynamo is an open-source framework for serving large models split across many GPUs and servers.

The largest gains Nvidia reported came from Samsung's chip engineering. Samsung is using Nvidia libraries to speed up the sparse linear calculations in its technology computer-aided design work, and its sMOL materials platform ran some quantum chemistry calculations up to five times faster from start to finish, Nvidia said. In one chip-scale thermal and mechanical analysis built with Nvidia's PhysicsNeMo framework, combining AI-predicted material properties, coarser grids and GPU acceleration cut simulation time from about 18 hours on a 32-core CPU to 41 seconds, a speed-up of roughly 1,600 times.

Kakao Mobility plans to develop a level 4 autonomous driving system on Nvidia's Hyperion reference platform, powered by the DRIVE AGX Thor computer, and to use Nvidia's Halos safety tools during development and validation. Initial work will centre on Kia purpose-built vehicles, combining Kakao Mobility's own driving software with experience it has gained operating in Seoul's Gangnam district.

LG Electronics and Nvidia said they had demonstrated LG's CLOiD robot carrying out manipulation and manufacturing tasks in a working factory, using Nvidia's Isaac GR00T robot models. The companies built a loop that runs from real-world data collection through Cosmos-based data augmentation, model training and on-robot computing on Jetson AGX Thor, to evaluation in simulation, which they plan to reuse for training factory robots.

On agents, Hyundai Motor Group is exploring Nvidia's NemoClaw blueprints and OpenShell runtime, which runs agents in a sandbox with policy controls over data and tool access. Korean security start-up AIM Intelligence is integrating its Starfort platform, which detects malicious instructions and data leakage, with OpenShell, and is weighing a preliminary proof of concept in the public energy sector. LG AI Research is also opening API access to K-EXAONE-2.5, a model built using Nvidia's Nemotron datasets, including a set of 7 million synthetic Korean personas grounded in official statistics.

Nvidia gave no financial terms for any of the agreements. The SK Telecom and Rebellions systems built on NVLink Fusion are expected to be ready for deployment in 2028.